Quick fix for memory lag in KoboldCPP when running local models

Hey everyone. I wanted to share a trick I found after struggling for days with my local setup crawling to a halt. When you run local weights for your companion and the context gets too big, things tend to lag badly. I tried adjusting the thread count and changing context window sizes inside the launcher GUI, but that did not really solve the core slowdown issue during long chat sessions. What finally worked for me was enabling BLAS acceleration properly and setting the smart memory allocation flags in the start script. It cut my response time in half and made memory retention way more stable. Has anyone else tried tweaking these specific flags yet? Let me know if you run into similar performance drops or if you found another workaround that helps keep conversations smooth.